Speech-Audio-Filter 🔊

Noise suppression in human speech audio samples using generative ML model. Identifying a person's audio from a mixed environment input.

Neural Network

• Encoder-Decoder
• Framework : TensorFlow(2.11.0)
• Dataset : https://datashare.ed.ac.uk/handle/10283/1942
  
  

Model Training

The architecture for a basic model is found in ./base_model.py. Use below code to create model.

model = create_model(batching_size, activation_func)

model.compile(optimizer, loss)
history = model.fit(train_dataset,epochs)

How to use:

• Clone repository to local machine and open predict.py
  
• Use function predict(audio_path, model_path).

Results

Input



Ground Truth



Prediction vs Input

🟠Prediction 🔵Input

To Do:

• Quantize model using TFLite